Knowledge-based region labeling for remote sensing image interpretation

EffCDNet: Transfer learning with deep attention network for change detection in high spatial resolution satellite images

Recent access to Very High Spatial Resolution (VHSR) Satellite Images allows vegetation monitoring at metric and sub-metric scale, with individual trees now detectable. Therefore, it discloses new applications in precision agriculture for orchards and other tree crops.

Assessment and monitoring of forest structure is frequently done with absolute density measures with field forest inventory data and expansion methods. The development of basal area and the number of trees estimation functions with data derived from very high spatial resolution satellite images enable their short-term and cost-effective evaluation, allowing also the estimation for the area not requiring extrapolation methods. The functions of basal area and the number of trees per hectare are based on crown cover obtained with very high spatial resolution satellite images for two evergreen oaks and umbrella pine. The three tree species are especially important in the agroforestry systems of the Mediterranean region. The linear functions fitted for pure stands of the three species and mixed stands of cork and holm oak and of cork oak and umbrella pine showed a better performance for basal area than for the number of trees per hectare. The inclusion of dummy variables for species composition improved the accuracy of the functions.

Best period for high spatial resolution satellite images for the detection of marks of buried structures

Landslide mapping (LM) has recently become an important research topic in remote sensing and geohazards. The area near the Three Gorges Reservoir (TGR) along the Yangtze River in China is one of the most landslide-prone regions in the world, and the area has suffered widespread and significant landslide events in recent years. In our study, an object-oriented landslide mapping (OOLM) framework was proposed for reliable and accurate LM from ‘ZY-3’ high spatial resolution (HSR) satellite images. The framework was based on random forests (RF) and mathematical morphology (MM). RF was first applied as an object feature information reduction tool to identify the significant features for describing landslides, and it was then combined with MM to map the landslides. Three object-feature domains were extracted from the ‘ZY-3’ HSR data: layer information, texture, and geometric features. A total group of 124 features and 24 landslides were used as inputs to determine the landslide boundaries and evaluate the landslide classification accuracy. The results showed that: (1) the feature selection (FS) method had a positive influence on effective landslide mapping; (2) by dividing the data into two sets, training sets which consisted of 20% of the landslide objects (OLS) and non-landslide objects (ONLS), and test sets which consisted of the remaining 80% of the OLS and ONLS, the selected feature subsets were combined for training to obtain an overall classification accuracy of 93.3% ± 0.12% of the test sets; (3) four MM operations based on closing and opening were used to improve the performance of the RF classification. Seven accuracy evaluation indices were used to compare the accuracies of these landslide mapping methods. Finally, the landslide inventory maps were obtained. Based on its efficiency and accuracy, the proposed approach can be employed for rapid response to natural hazards in the Three Gorges area.

Very high spatial resolution (VHR) satellite images possess several advantages in terms of describing the details of ground targets. Extracting built-up areas from VHR images has received increasing attention in practical applications, such as land use planning, urbanization monitoring, geographic information database update. In this study, a novel method is proposed for built-up area detection and delineation on VHR satellite images, using multi-resolution space-frequency analysis, spatial dependence modelling and cross-scale feature fusion. First, the image is decomposed by multi-resolution wavelet transformation, and then the high-frequency information at different levels is employed to represent the multi-scale texture and structural characteristics of built-up areas. Subsequently, the local Getis-Ord statistic is introduced to model the spatial patterns of built-up area textures and structures by measuring the spatial dependence among frequency responses at different spatial positions. Finally, the saliency map of built-up areas is produced using a cross-scale feature fusion algorithm, followed by adaptive threshold segmentation to obtain the detection results. The experiments on ZY-3 and Quickbird datasets demonstrate the effectiveness and superiority of the proposed method through comparisons with existing algorithms.

Analytical 3D views and virtual globes — scientific results in a familiar spatial context

Deep learning (DL) convolutional neural networks (CNNs) have been rapidly adapted in very high spatial resolution (VHSR) satellite image analysis. DLCNN-based computer visions (CV) applications primarily aim for everyday object detection from standard red, green, blue (RGB) imagery, while earth science remote sensing applications focus on geo object detection and classification from multispectral (MS) imagery. MS imagery includes RGB and narrow spectral channels from near- and/or middle-infrared regions of reflectance spectra. The central objective of this exploratory study is to understand to what degree MS band statistics govern DLCNN model predictions. We scaffold our analysis on a case study that uses Arctic tundra permafrost landform features called ice-wedge polygons (IWPs) as candidate geo objects. We choose Mask RCNN as the DLCNN architecture to detect IWPs from eight-band Worldview-02 VHSR satellite imagery. A systematic experiment was designed to understand the impact on choosing the optimal three-band combination in model prediction. We tasked five cohorts of three-band combinations coupled with statistical measures to gauge the spectral variability of input MS bands. The candidate scenes produced high model detection accuracies for the F1 score, ranging between 0.89 to 0.95, for two different band combinations (coastal blue, blue, green (1,2,3) and green, yellow, red (3,4,5)). The mapping workflow discerned the IWPs by exhibiting low random and systematic error in the order of 0.17–0.19 and 0.20–0.21, respectively, for band combinations (1,2,3). Results suggest that the prediction accuracy of the Mask-RCNN model is significantly influenced by the input MS bands. Overall, our findings accentuate the importance of considering the image statistics of input MS bands and careful selection of optimal bands for DLCNN predictions when DLCNN architectures are restricted to three spectral channels.

New ongoing rural construction has resulted in an extensive mixture of new settlements with old ones in the rural areas of China. Understanding the spatial characteristic of these rural settlements is of crucial importance as it provides essential information for land management and decision-making. Despite a great advance in High Spatial Resolution (HSR) satellite images and deep learning techniques, it remains a challenging task for mapping rural settlements accurately because of their irregular morphology and distribution pattern. In this study, we proposed a novel framework to map rural settlements by leveraging the merits of Gaofen-2 HSR images and representation learning of deep learning. We combined a dilated residual convolutional network (Dilated-ResNet) and a multi-scale context subnetwork into an end-to-end architecture in order to learn high resolution feature representations from HSR images and to aggregate and refine the multi-scale features extracted by the aforementioned network. Our experiment in Tongxiang city showed that the proposed framework effectively mapped and discriminated rural settlements with an overall accuracy of 98% and Kappa coefficient of 85%, achieving comparable and improved performance compared to other existing methods. Our results bring tangible benefits to support other convolutional neural network (CNN)-based methods in accurate and timely rural settlement mapping, particularly when up-to-date ground truth is absent. The proposed method does not only offer an effective way to extract rural settlement from HSR images but open a new opportunity to obtain spatial-explicit understanding of rural settlements.

A change detection in multi-temporal Very High Spatial Resolution (VHR) Satellite Images (SI) has a great importance in land survey and to document land change, residential areas changes and damage assessment caused by crises and natural disasters. We present a novel approach based on Self-Organizing Map (SOM) Artificial Neural Networks (ANN) to detect changes in multitemporal VHR SI based upon object level representation as a spatial vector and not only upon pixel level of image, from one hand. On other hand, identify of new types of changes, such as: Expanded Object (ExO), Erosion Object (ErO), Moved Object (MO), and not only identify as Stable Object (SO), Disappeared Object (DO), New Object (NO) as in most previous studies. We will explain in this way an application of the proposed method on VHR SI from two satellites QuickBird and Worldview-2 to detect and identify changes types in random residential area near Aleppo Citadel during Syrian Crisis period (from 2011 to 2016 for Aleppo City). The results and rates of correct changes detection are high, and the maps showed many discovered changes using our method.

Accurate information on the location and magnitude of vegetation change in scenic areas can guide the configuration of tourism facilities and the formulation of vegetation protection measures. High spatial resolution remote sensing images can be used to detect subtle vegetation changes. The major objective of this study was to map and quantify forest vegetation changes in a national scenic location, the Purple Mountains of Nanjing, China, using multi-temporal cross-sensor high spatial resolution satellite images to identify the main drivers of the vegetation changes and provide a reference for sustainable management. We used Quickbird images acquired in 2004, IKONOS images acquired in 2009, and WorldView2 images acquired in 2015. Four pixel-based direct change detection methods including the normalized difference vegetation index difference method, multi-index integrated change analysis (MIICA), principal component analysis, and spectral gradient difference analysis were compared in terms of their change detection performances. Subsequently, the best pixel-based detection method in conjunction with object-oriented image analysis was used to extract subtle forest vegetation changes. An accuracy assessment using the stratified random sampling points was conducted to evaluate the performance of the change detection results. The results showed that the MIICA method was the best pixel-based change detection method. And the object-oriented MIICA with an overall accuracy of 0.907 and a kappa coefficient of 0.846 was superior to the pixel-based MIICA. From 2004 to 2009, areas of vegetation gain mainly occurred around the periphery of the study area, while areas of vegetation loss were observed in the interior and along the boundary of the study area due to construction activities, which contributed to 79% of the total area of vegetation loss. During 2009–2015, the greening initiatives around the construction areas increased the forest vegetation coverage, accounting for 84% of the total area of vegetation gain. In spite of this, vegetation loss occurred in the interior of the Purple Mountains due to infrastructure development that caused conversion from vegetation to impervious areas. We recommend that: (1) a local multi-agency team inspect and assess law enforcement regarding natural resource utilization; and (2) strengthen environmental awareness education.

Government agencies responsible for riparian environments are assessing the combined utility of field survey and remote sensing for mapping and monitoring indicators of riparian zone health. The objective of this work was to determine if the structural attributes of savanna riparian zones in northern Australia can be detected from commercially available remotely sensed image data. Two QuickBird images and coincident field data covering sections of the Daly River and the South Alligator River – Barramundie Creek in the Northern Territory were used. Semi-variograms were calculated to determine the characteristic spatial scales of riparian zone features, both vegetative and landform. Interpretation of semi-variograms showed that structural dimensions of riparian environments could be detected and estimated from the QuickBird image data. The results also show that selecting the correct spatial resolution and spectral bands is essential to maximize the accuracy of mapping spatial characteristics of savanna riparian features. The distribution of foliage projective cover of riparian vegetation affected spectral reflectance variations in individual spectral bands differently. Pan-sharpened image data enabled small-scale information extraction (<6 m) on riparian zone structural parameters. The semi-variogram analysis results provide the basis for an inversion approach using high spatial resolution satellite image data to map indicators of savanna riparian zone health.

Mapping local density of young Eucalyptus plantations by individual tree detection in high spatial resolution satellite images

Identifying concentrated areas of trip generators from high spatial resolution satellite images using object-based classification techniques

Monitoring mangrove forests after aquaculture abandonment using time series of very high spatial resolution satellite images: A case study from the Perancak estuary, Bali, Indonesia